Segmentation is one of the old concepts being used across fields and by non-technical people. People are segmented at least into two groups “Male” and “Female” based on gender characteristics. Now let’s try to understand these two segments better. These two segments will be very different on a few characteristics (that are variables in analytics definition) such as vocal (e.g. Pitch of voice), physical (e.g. height) and hormonal. The segments may not vary different based on other characteristics such as age, education level and success in various fields.
How can leverage the above segmentation understanding?
We can look at the images and classify them into two segments – “Male” and “Female”. Based on characteristics of the picture, machine learning & statistical techniques can be used for building classification or segmentation model for tagging new images into these two segments. These models can be implemented on the servers for various checks & decisions.
Other important point to note is that segmentation scheme is always linked to a list of characteristics. If the list of characteristics changes, the segmentation scheme will be different - meaning the different segments will be created.
What is type of the segmentation class? Since the images or people can be classified into these two groups, the labels are available. So this is an example of supervised or objective segmentation.
Consider another example, people from geographic locations visit different type of retailers for their grocery shopping. There is no condition which suggests that a specific type of people should visit only specific type of retailers. Having said that, the retailers and government may be interested to know who typically visit which type of retailers.
Now based on spend level across different type of retailers – supermarket, hypermarket, departmental store, convenience store, Kirana stores and online retailer– segments can be created to understand different shopping patterns across customers.
In this example, before segmentation approach is executed, there is no understanding of number of segments to be created and what level of spend across retailer types to be used for segment creation. In other words, there is not data label; hence this is a subjective or unsupervised segmentation class.
Once you create these segments, steps are taken to understand why customers are behaving in a particular way or spending higher on specific type of retail stores. This is typically called segment profiling and labeling. Based on a specific business objective, an appropriate segment based strategy will be developed for achieving business objective.
One of the commonly used Subjective Segmentation techniques is K-means clustering. Based on a list of input variables and maximum clusters to be created, the K means clustering algorithm identify the segments or clusters.